Correspondence should be addressed to Saija Mauno, Department of Psychology, University of Jyväskylä, P.O. Box 35, 40014, University of Jyväskylä, Finland (e-mail: firstname.lastname@example.org.
This study examined whether perceived work–family conflict would function as a mediator in the link between work–family culture perceptions and self-reported distress. Data were obtained from employees (N=1,297) of five Finnish organizations representing both the public (local social and health care, school, and labour departments) and the private sectors (paper mill, IT company). The results showed that perceived work–family conflict functioned as a partial mediator between employees' perceptions of work–family culture and self-reported distress in two organizations (i.e. in the social and health care department and paper mill), whereas the relationship turned out to be direct in the other three organizations (i.e. the education, labour departments and IT company). Thus, a supportive work–family culture was related directly and indirectly, through reduced work–family conflict, to the well-being of employees.
Recently, a new concept, work–family culture, has been introduced in the work–family literature (e.g. Campbell Clark, 2001; Kinnunen, Mauno, Geurts, & Dikkers, 2005; Lewis, 2000; Lewis & Smithson, 2001; Thompson, Beuvais, & Lyness, 1999). Generally, work–family culture refers to an organization's supportiveness or responsiveness towards employees' family-related needs. It has also been recognized that a supportive work–family culture is associated with several positive well-being outcomes (see Kinnunen et al., 2005, for a review). However, the mechanism that explains these positive relationships remains unclear. Specifically, the issue that needs further clarification is whether the relationship between a supportive work–family culture and well-being is direct or mediated by other factors. In the present study, we aimed to shed further light on this issue.
On the basis of previous (mainly theoretical) work (Kinnunen et al., 2005; Lewis, 2000; Lewis & Smithson, 2001), the basis of the above-mentioned mechanism is as follows: where a non-supportive work–family culture exists, employees may not feel entitled to use the work–family arrangements that are available to them (see Lewis & Smithson, 2001) and thus may not make full use of them. In fact, studies also provide some empirical evidence that a supportive family–work culture increases the uptake of work–family benefits, for example, flexitime arrangements, part-time work, a compressed working week, and family care leave (see Allen, 2001; Dikkers, Den Dulk, Geurts, & Piper, 2005; Haas, Allard, & Hwang, 2002; Thompson et al., 1999). However, the view that the uptake of work–family arrangements leads to a lower level of work–family conflict, and thus to a better level of well-being, is mainly based on theoretical reasoning. That is, if employees do not feel entitled to take advantage of existing work–family arrangements, then they may encounter even more problems in balancing the demands of work and family and experience a decreased level of well-being (e.g. Lewis & Smithson, 2001). Therefore, there are at least two potential mediators – the use of work–family arrangements and the perceived work–family conflict – which may explain the positive outcomes of a supportive work–family culture on well-being.
Of the two potential mediators, we focused on perceived work–family conflict in the present study. The reasons for this are twofold. First, despite its apparent theoretical soundness, work–family conflict has received little empirical attention as a potential mediator between work–family culture and well-being. Indeed, it has recently been proposed that perceived work–family conflict is potentially an even more important mediator variable than objective factors such as the availability of work–family arrangements (see Eby, Casper, Lockwood, Bourdeaux, & Brinley, 2005). Second, there are country-specific factors that play an important role in this issue. In Finland, many work–family arrangements have statutory force (day-care and other care services for those who need them, paid parental leave, maternity and paternity leave, leave for the care of a sick child, leave for the care of young children), or are based on other official agreements (there is a wide array of working time arrangements, including flexible working hours). Hence, their uptake might not be as dependent on the prevailing organizational work–family culture as in many other countries. On the contrary, in this specific context, work–family conflict as perceived by individual employees is likely to assume greater importance.
In the present study, we investigated whether perceived work–family conflict would operate as a mediating factor between work–family culture perceptions and self-reported distress. Furthermore, our multiorganizational data (N=1,297) enabled us to study simultaneously whether the relationships were equivalent across five Finnish organizations (i.e. the social and health care department, education department, labour department, paper mill, and IT company). In our model, in which testing was performed using structural equation modelling (SEM), the fully mediating model was taken as the starting-point (see Fig. 1).
Direct relationship between work–family culture and well-being
Recently, Thompson et al. (1999) defined work–family culture as ‘the shared assumptions, beliefs and values regarding to the extent to which an organization supports and values the integration of employees’ work and family lives' (p. 349). Central to work–family culture is the construct of supportiveness, which generally refers to the extent that an organization (or work unit/department) is perceived to be family-supportive by its personnel (e.g. Allen, 2001; Dikkers et al., 2005; Haas et al., 2002; Kinnunen et al., 2005). For example, Allen discusses family-supportive organization perceptions (FSOP), which include the organization's supportiveness toward the demands of its employees' families.
In examining the links between work–family culture and well-being outcomes, the theoretical assumption is that a supportive work–family culture should make an organization a more pleasant place to work in. This, in turn, should affect an employee's work experiences, including work–family interface, positively (see Allen, 2001; Behson, 2002; Casper & Buffardi, 2004). A supportive organizational culture should signal to employees that the organization is willing to look after the well-being of its personnel (e.g. Cameron & Quinn, 1999; Goodman, Zammuto, & Gifford, 2001). The theoretical basis for the above argumentation can be found from perceived organizational support theory (i.e. POS theory; Eisenberger, Huntington, Hutchison, & Sowa, 1986; Rhoades & Eisenberger, 2002). According to the POS theory, an organization that values the well-being of its employees by showing concern for their needs, goals, and personal problems, also promotes the well-being and health of its personnel.
Indeed, previous empirical studies confirm that a supportive work–family culture is associated with several positive outcomes. For example, it has been found that when the personnel perceives an organization's culture as family supportive, they report a lower level of psychological distress (Kossek, Colquitt, & Noe, 2001; Mauno, Kinnunen, & Piitulainen, 2005). Regarding work attitudes, it has been indicated that supportive work–family culture fosters job satisfaction (Allen, 2001; Behson, 2002; Mauno et al., 2005) and organizational commitment (Dikkers et al., 2005; Lyness, Thompson, Francesco, & Judiesch, 1999; Thompson et al., 1999). In sum, there is a strong theoretical and empirical basis for a positive association between supportive work–family culture and employee well-being. Therefore, we hypothesized that as a specific form of organizational support, a family-supportive organizational culture (i.e. work–family culture) would relate positively to the self-reported well-being of employees.
Mediating role of work–family conflict
The empirical evidence obtained on the potential mediating role of work–family conflict in the link between work–family culture and employee well-being has mostly been indirect. First, it has been shown that a family supportive organizational culture is linked to a low level of work–family conflict (Dikkers et al., 2005; Mauno et al., 2005; Thompson et al., 1999). Second, there are also research findings suggesting that work-related social support, which is conceptually similar to the concept of supportive work–family culture, operates as an antecedent to work–family conflict (Carlson & Perrewé, 1999; Warren & Johnson, 1995). Third, there is a lot of evidence to show that work–family conflict is linked to various indicators of diminished individual well-being (see Allen, Herst, Bruck, & Sutton, 2000, for a review). This evidence, together with the aforementioned support for a direct relationship between supportive work–family culture and employee well-being, points to the possibility of mediation.
In this respect, a recent Dutch study provides some preliminary findings, showing that work–family conflict partially mediated the association between a supportive work–family culture and job satisfaction (Peeters, Montgomery, & Schaufeli, 2003). Furthermore, Thomas and Ganster (1995) have previously found that supervisor support – as one of the most important determinants of a supportive work–family culture – reduced work–family conflict, which, in turn, was associated with increased job satisfaction and decreased depression and somatic complaints. Moreover, Snow, Swan, Raghavan, Connell, and Klein (2003) recently indicated that a lack of work-related social support (including low supervisor support) increased work–family conflict, which, in turn, resulted in increased psychosomatic symptoms (anxiety, depression, somatic complaints). Taken together, these studies provide empirical evidence that work–family conflict may function as a mediating factor between work- or organization-related support (defined and measured in our study through work–family culture) and employee well-being.
At the theoretical level, the mediator model developed by Frone et al. (1992, 1997), which states that perceived work–family conflict operates as a mediating mechanism between work and family demands and employee well-being in different domains of life, leads us to a similar conclusion. That is, a non-responsive work–family culture may also be viewed as an organizational demand or constraint, which starts by exacerbating perceived work–family conflict, and then negatively spills over into lower self-reported well-being in different domains of life. Consequently, considering both empirical and theoretical viewpoints, our second hypothesis was that the positive link between supportive work–family culture and self-reported well-being would disappear when perceived work–family conflict is taken into account. This highlights that the relationship between work–family culture and experienced well-being is mediated by perceived work–family conflict.
However, the previous studies in which the mediating role of work–family conflict in the relationship between work–family culture and employee well-being has been examined have certain limitations. For example, in the study of Peeters and associates (2003), outcome variables consisted solely of job attitudes, whereas in the study conducted by Thomas and Ganster (1995), work–family culture was measured unidimensionally, that is, through supervisor support (see also Snow et al., 2003). Our study extends this research by focusing on outcome variables other than job attitudes (i.e. self-reported distress) and by conceptualizing work–family culture more broadly (i.e. multidimensionally). Furthermore, our study is based on five diverse organizational samples, thus enhancing the generalizability of our findings. Taken together, our study clearly contributes to the previous literature and research.
Participants and procedure
The data for the study were obtained between the years 2001 and 2003 by means of a questionnaire from a sample of men and women (N=1,297) employed in five organizations located in central Finland. The organizations were drawn from both the public (N=914) and the private (N=383) sectors of the economy. The public sector organizations, which were female-dominated, were a local social and health care department (men N=25, women N=471, response rate 59%), education department (men N=70, women N=162, response rate 79%), and labour department (men N=40, women N=142, response rate 76%). The private-sector organizations, which were male-dominated, were an information technology (IT) company (men N=119, women N=80, response rate 51%) and a paper mill (men N=144, women N=39, response rate 62%).
The social and health care department provides health and social services to a town of some 80,000 inhabitants. Because the organization is relatively large (with over 2,000 employees), sampling was carried out in units of less than 50 employees; all the employees participated in the study, whereas in the larger units, a random sample was selected. The main occupational groups were nurses, doctors, dentists, child-care personnel, and social workers. The education department employs nearly 300 people and provides primary and lower secondary education for children aged between 7 and 16 years in a rural community with a population of about 15,000. All the employees, among whom teachers formed the largest occupational group, were asked to take part in the study. The data from the labour department were gathered in February 2003, 1.5 years later than the data in the other organizations. The local labour department, which has 241 employees, is administered by the Ministry of Labour and it supports the regional development of the labour market, providing a variety of services both for employees (employment search and career counselling services, and training, e.g. vocational courses) and employers (recruiting and selection services, e.g. job advertisements) in central Finland. The main occupational groups were employment authorities, employment consultants, and vocational guidance psychologists.
The paper mill employs nearly 300 people and is part of a larger group of companies in the forestry industry. It specializes in pulp for fine papers and folding boxboard, and the main occupational group consisted of (male) blue-collar workers (e.g. machine operators, shift supervisors). Similarly, the IT company is part of a larger group of companies. It produces applications and software for customers in both the private and public sectors. Of the over 400 people employed in the IT company, most were senior and managerial employees (e.g. ADP analysts, specialists, and project managers). In both of the private sector organizations, all the employees were asked to participate in the study. The paper industry, which has a long history, and the more recent information technology companies together represent important segments of the Finnish economy.
Table 1 presents the demographic characteristics of the participants by organization. As mentioned already, gender segregation was clearly prevalent in the organizations. The vast majority of the respondents in each organization were aged between 35 and 54 years, although the employees in the IT company turned out to be younger than those in the other organizations. The participants' educational level seemed to vary by organization. Employees at the paper mill and in the social and health care department had the least higher education levels, whereas higher education was more common in the education and labour departments as well as in the IT company. For other background factors, see Table 1.
Table 1. Characteristics of study participants by organizations (F, %)
Work-family culture was measured via three scales, which we adapted from Thompson et al. (1999). These three scales assess family supportiveness of management, career development, and working hours. The items in each scale were assessed on a 7-point scale (1 = strongly disagree, 7 = strongly agree). On each scale, higher values reflected a more supportive work–family culture. Cronbach alpha coefficients, means, and standard deviations for the computed sum scales are presented in Table 2.
Table 2. Cronbach α coefficients, means, standard deviations and zero-order correlations for the whole sample
First, the Management culture scale rated the extent to which managers were perceived to be supportive and sensitive to the employees' family responsibilities. We used 5 of the original 11 items (see Allen, 2001) referring directly to managers or management (e.g. ‘In general, managers in this organization are quite accommodating to family-related needs’, ‘Middle managers and executives in this organization are sympathetic toward employees' child care responsibilities’).
Second, the Career development culture scale measured the extent to which career consequences were perceived to follow work–family arrangements, such as parental leave, or turning down tasks for family-related causes. The scale contained five items (e.g. ‘Many employees are resentful when men in this organization take extended leave to care for new-born or adopted children [reverse scored]’, ‘In this organization, employees who participate in the available work–family programmes, for example, job sharing, part-time work, are viewed as less serious about their careers than those who do not participate in these programmes [reverse scored]’).
Third, the Working hours culture scale measured the extent to which the employees experienced organizational time demands or expectations that might impede the work–family interface, and contained four items (e.g. ‘To get ahead in this organization, employees are expected to work more than 50 hours a week, whether in the workplace or at home [reverse scored]’, ‘Employees are expected to take work home at night and/or weekends [reverse scored]’).
Work–family conflict was assessed through four items from the Pressure Management Indicator Questionnaire (PMI; see Williams, 1996; Williams & Cooper, 1998). The items referred to negative work spillover into family life and the respondents were asked for to evaluate whether these matters (e.g. ‘not being able to “switch off” at home’, ‘demands my work makes on my relationship with my partner or children’) had been sources of pressure or stress at work during the past 3 months, and a 6-point rating scale was used (1 = definitely is not a source of stress, 6 = definitely is a source of stress). Thus, a high score indicates a high degree of work–family conflict. It should be noted that the Work–family conflict scale used in this study is slightly different from many other work–family conflict scales (cf. Carlson, Kacmar, & Williams, 2000; Frone et al., 1992). However, it has been shown previously to have good validity and reliability in several Finnish samples (Piitulainen, Mauno & Kinnunen, 2002a, 2002b).
All the dependent variables describing the level of employee well-being, physical symptoms, exhaustion, and negative job-related mood were also measured using the scales from the PMI (see Williams, 1996; Williams & Cooper, 1998). The Physical symptoms scale consisted of three items (e.g. ‘shortness of breath or feeling dizzy’, ‘muscles trembling, e.g. eye twitch’) and the Exhaustion scale consisted of four items (e.g. ‘feeling unaccountably tired or exhausted’, ‘feeling as though you do not want to get up in the morning’). Both scales assess general health-related stress symptoms, which are often measured in occupational stress research (e.g. Spector & Jex, 1998; Warr, 1990; Williams & Cooper, 1997). Specifically, the respondents were requested to evaluate how often they have felt these symptoms over the last 3 months (1 = never, 6 = very frequently).
The Negative job-related mood scale consisted of five items linked to negative feelings experienced at work, such as irritation, anxiety, and lack of self-confidence (e.g. ‘During an ordinary working day, are there times when you feel unsettled and upset though the reasons for this might not always be clearly obvious?’, ‘Are there times at work when you feel so exasperated that you sit back and think to yourself that life is really too much of an effort?’, and, ‘As you do your job, have you noticed yourself questioning your own ability and judgment and a decrease in your overall self-confidence?’). The respondents were asked to evaluate whether they have had feelings and experiences of this type because of their job pressures over the last 3 months (1 = very untrue, 6 = very true).
In addition, the following background factors were studied as control (covariates) variables (see Table 1): gender (1 = men, 2 = women), age (1 = 18–34, 2 = 35–44, 3 = 45–54, 4 = 55–64 years), and higher education (1= none or short courses, 2 = vocational school, 3 = vocational college, 4 = university). Factors of this type have also been used as control variables in previous work–family culture studies (see e.g. Allen, 2001; Campbell Clark, 2001; Mauno et al., 2005; Thompson et al., 1999), and therefore their effects were controlled for in SEM analysis. In particular, gender is an important background factor in work–family research (e.g. Eby et al., 2005), whereas age and education need to be controlled for because they often relate to distress outcomes (e.g. Birdi, Warr, & Oswald, 1995; Dear, Henderson, & Korten, 2002; Murray, Gien, & Solberg, 2003). The latter, in fact, formed the dependent variables in the present study. Zero-order correlations between the study variables for the whole sample are presented in Table 2.
SEM, performed by the M-Plus programme (Muthén & Muthén, 1998), was used to test the hypothesized mediator model (Fig. 1). The programme is a relatively new statistical software package appropriate for various correlative analyses, including SEM. We estimated the model via multiple group analysis (MGA), which is a highly recommended method of studying whether the chosen parameters are equal across different groups (i.e. across organizations in the present study). MGA uses the full data matrix but simultaneously allows group-specific modifications as necessary. The covariance matrices served as the data input form, and the method of estimation was maximum likelihood. It should, however, be noted that if latent constructs (measurement models) are also included in the MGA, then the programme automatically constrains the measurement parameters of the model to be equal across groups. Fit statistics and parameter values indicate whether the model is adequate for the groups or whether it needs group-specific modifications. However, in the MGA analysis, the parameters of the residual variances of the observed dependent variables as well as structural parameters (beta parameters) are freely estimated across groups.
In line with the MGA procedure (see Muthén & Muthén, 1998), we first estimated whether the hypothesized mediating model (see Fig. 1) would hold for each organization. Where necessary, separate models for each organization were then specified on the basis of modification indices and t values. Accordingly, parameters having modification indices of 8.0 or more were included in the model, and then those paths whose t values remained below 2.0 were deleted from the model, even if they would have been included in our hypothesized model.
The goodness-of-fit of the models was evaluated against the root mean square error of approximation (RMSEA; should be <.05), standardized root mean square residual (SRMR; should be <.05), comparative fit index (CFI; should be >.90) and Tucker–Lewis index (TLI; should be >.90; see Hu & Bentler, 1999, for cut-off criteria of fit indices). A frequently used fit index, the chi-squared statistic, could not be utilized because of its high sensitivity to large sample sizes (Bagozzi & Yi, 1988). However, we were able to use the chi-squared test in the model comparison, which was also conducted on the basis of Akaike's information criterion (AIC), Baylesian BIC index, and sample-size adjusted BIC (ABIC) index: the smaller the values, the better the model. Altogether, model goodness was evaluated through parameter values (t values >2.0), modification indices, goodness of fit statistics, model parsimony, and its theoretical soundness.
It should be noted that in the SEM framework, a mediator effect is shown by a statistically significant t value (>2.0) for an estimated parameter. In the present case, this means that if work–family conflict operates as a fully mediating factor between work-family culture and self-reported distress, then the t value for a structural link between work–family culture and distress (i.e. the direct link between independent and dependent variable) should be non-significant (i.e. less than 2.0) when indirect links (i.e. a path between work–family culture and work–family conflict, and a path between work–family conflict and distress) are estimated. Such a finding is consistent with the test for fully mediation suggested by Baron and Kenny (1986). In partial mediation, t values for direct and indirect relations are both significant (i.e. >2.0), and should therefore be included in the model. If partial mediation appears, all structural links between the studied phenomena (here work–family culture, work–family conflict, distress) should be significant in terms of t values. This test of partial mediation also fulfils the conditions mentioned by Baron and Kenny for partial mediation.
The preliminary correlative analysis (see) showed that the correlations between the work–family culture variables (management culture, career development culture, working hours culture) were relatively high across organizations, ranging from .34 to .73. In addition, the intercorrelations between the outcome variables (physical symptoms, exhaustion, negative job-related mood) were relatively robust across organizations, ranging from .31 to .62. Thus, in order to avoid problems related to multicollinearity in our model analysis, we first constructed latent variables and explored their equivalence across organizations.
The first latent scale was labelled the supportiveness of work–family culture and included three sum scales (i.e. management culture, career development culture and working hours culture). The second latent scale was labelled the self-reported distress and contained three sum scales (i.e. physical symptoms, exhaustion and negative job-related mood). The standardized factor loadings for both the latent scales were all significant across organizations. For the work–family culture scale, the factors loadings were .53, .68, and .62, respectively. For the self-reported distress scale, the factor loadings were .64, .86, and .74, respectively.
Fit indices indicated that the model showed a good fit with the data: CFI = .985, TLI = .980, RMSEA = .042, SRMR = .049. Consequently, we received a satisfactory equivalent factor structure for the scales across organizations.
First, we estimated the hypothesized structural (or beta) parameters of the fully mediating model (Model A, Table 3; see also Fig. 1). However, modification indices (above the criterion value 8.0) suggested a direct link between work–family culture and self-reported distress for each organization (MI ranged from 11.723 to 24.119 for organizations). Thus, we modified the first model by freely estimating this direct path across organizations.
Table 3. Fit and model comparison indices for the estimated models
This modification resulted in better fit and model comparison indices (a decrease in chi-square, AIC, BIC, and ABIC) than were received for the hypothesized fully mediating model (cf. Steps 1 and 2 in Table 3). Thus, the partially mediating model (Model B, Table 3), turned out to be superior to the hypothesized fully mediating model (Model A). The re-estimated path between work–family culture and self-reported distress was significant (t value > 2.0) for each organization (t values ranged from −3.009 to −4.763), suggesting that the direct path should be included into the model.
After this modification, we re-examined the t values for all the estimated parameters across the organizations, and noticed that the path between work–family culture and work–family conflict was non-significant (t value < 2.0) for the education and labour departments. Accordingly, we omitted this path for these two organizations, resulting in another modified model (Model C; see Step 3 in Table 3). The re-estimated model showed adequate fit with the data and a chi-squared comparison test for the nested models (χ2 difference test, df(2)=3.052) indicated that deleting the non-significant path did not impair the model. The other smaller model comparison values (AIC, BIC, ABIC) also suggested that re-estimated Model C (Step 3) was better than Model B (Step 2). Furthermore, Model C had a benefit of greater parsimony. Thus, the direct model was supported in the education and labour departments.
The role of covariates
We included the covariates (gender, age, education) in the model in order to examine whether, after taking them in account, the above effects would remain significant in explaining the studied phenomena (i.e. work–family culture, work–family conflict, self-reported distress). The fit indices showed sufficient fit (see Step 4 in Table 3). However, there were many non-significant effects (t values < 2.0) for the covariates across the organizations. These effects were then removed from the model (see Step 5 in Table 3). However, it should be noted that when the covariates were included in the model, the effect between work–family culture and work–family conflict became non-significant for the IT company. Thus, we deleted this parameter as well. After these modifications, we re-estimated the model. This resulted in a better fit according to the model comparison indices (cf. Steps 4 and 5). The effects of covariates are described below.
Because the structural parameters between the work–family conflict and self-reported distress (coefficients .43, .37, .42, .42, .52 for the five organizations, respectively) as well as the path between work–family culture and self-reported distress (coefficients −.28, −.32, −.31, −.39, −.30, respectively) seemed to be relatively equal across organizations, we constrained these parameters to be equal across organizations. Consequently, this was conducted as a final step in our model modification process (see Step 6, Table 3). Each fit index showed again a good fit for the constrained model. Specifically, the models were compared using the chi-squared difference test, which indicated that these structural paths could be constrained equal across organizations: The chi-square diference test, df(8)=3.433. If the non-constrained model had been superior to the constrained model, then the difference between chi-squared values should have exceeded 15.507. Moreover, each model comparison index shows a decrease, supporting the conclusion that the constrained model was superior to the unconstrained model. The final results of the MGA analysis (Step 6, Table 3) are summarized in Fig. 2 and reported in detail below.
The final model
As Fig. 2 shows, perceived work–family conflict partially mediated the link between work–family culture and self-reported distress in only two organizations; the social and health care department and in the paper mill. Hence, in these two organizations, an unsupportive work–family culture was linked to increased work–family conflict, which, in turn, resulted in increased self-reported distress. There was also a direct link between an unsupportive work–family culture and self-reported distress.
Figure 2 further reveals that in the other three organizations work–family culture perceptions were directly related to self-reported distress. Furthermore, it is important to note that this association was constrained to be equal across organizations. Although the beta coefficients are not exactly the same, they do not vary across the organizations in statistical terms (coefficients are marked by asterisks in Fig. 2). The proportions of variance for the dependent variable (distress) explained by the factors in the model were relatively high, ranging from 25% (the education department) to 41% (the IT company; see Fig. 2).
The covariates included in the model showed associations which varied across organizations. First, gender was associated with work–family conflict in the paper mill (β=−0.27) and in the labour department (β=−0.16) where men more often reported work–family conflict than women did. Second, age was related to each phenomenon examined. In the social and health care (β=−0.12) and education (β=−0.17) departments, younger employees perceived their organization's work–family culture more positively than the older personnel did. In the education department, older respondents experienced work–family conflict more often than did their younger counterparts (β=0.19). Self-reported distress was found to increase along with advancing age in the social and health care department (β=0.10) and in the IT company (β=0.29). Third, the level of education turned out to be a significant correlate, in particular in the education department, where high education was related to perceptions of supportive work–family culture on the one hand (β=0.41), and to experiences of work–family conflict on the other hand (β=0.37). This latter path was also valid in the social and health care department (β=0.16).
To sum up the most important findings, our analyses failed to support the hypothesized fully mediating model suggested in Fig. 1. Instead, the partially mediating model received empirical support in two organizations; the social and health care department and in the paper mill. It also emerged that work–family culture was most often directly linked to distress without any mediation. The direct-effect model was valid in three organizations; the education department, IT company, and labour department.
Work–family culture perceptions and distress
Consistent with both our expectations and previous studies (see Allen, 2001; Behson, 2002; Campbell Clark, 2001; Kossek et al., 2001; Lyness et al., 1999; Thompson et al., 1999), a family-friendly organizational culture was associated with employees' well-being in terms of lower levels of self-reported symptoms of distress. This finding was valid for each of the five organizations studied, thus providing cross-validation evidence concerning the direct linkages between perceived work–family culture and employees' well-being. This finding also lends support to the POS theory (see Behson, 2002; Casper & Buffardi, 2004; Eisenberger,et al., 1986; Rhoades & Eisenberger, 2002), which argues that an organization's responsiveness towards employees' needs and values, including those related to work–family issues, should be taken into consideration in order to foster employees' well-being. Moreover, there is also empirical evidence that supports the assumption that perceived organizational support does promote employees' well-being, work attitudes, and health (see Behson, 2002; Rhoades & Eisenberger, 2002; Rhoades, Eisenberger, & Armeli, 2001). Consequently, although our study focused on a specific form of perceived organizational support (i.e. a perceived family–friendliness of an organization) our findings are consistent with those of previous research.
Nevertheless, in two organizations (the social and health care department and paper mill), the association between work–family culture perceptions and distress was partially mediated by experiences of work–family conflict. Hence, in these organizations, an unsupportive work–family culture resulted in increased work–family conflict, leading to increased distress (see also Peeters et al., 2003; Thomas & Ganster, 1995). At the same time, work–family culture perceptions were also directly linked to distress. However, it should also be noted that the hypothesized fully mediated model did not receive support in any organization.
The different relationships in the organizations detected in SEM analysis might be explained via a threshold effect. It is possible that there is a critical threshold below which work–family culture perceptions are not associated with perceived work–family conflict. If work–family culture perceptions are not robust enough (in either a positive or negative direction), then they are not associated with perceived work–family conflict. Applied to the present study, this assumption would suggest that such a threshold had been reached in the social and health department and in the paper mill, but not in the other three organizations. In fact, previous mean comparison analysis for the same data has shown that the most negative work–family culture perceptions were usually reported in these two organizations (Mauno et al., 2005; Mauno & Pyykkö, 2004).
In addition, perceived work–family conflict was operationalized somewhat differently compared with the other commonly used work–family conflict scales (e.g. Carlson et al., 2000; Frone et al., 1992), which have produced more robust associations between work–family culture perceptions and work–family conflict than found in the present study (e.g. Allen, 2001; Dikkers et al., 2005; Thompson et al., 1999). Therefore, the relatively modest or non-existent association found between work–family culture and work–family conflict may be connected to the differences in the measurement of work–family conflict. For example, it is possible that the other work-family conflict scales capture the experiences of work–family conflict more effectively than does the measure used in our study. Consequently, these scales may also have more predictive (criterion) validity with the respect to the antecedents (e.g. work–family culture) and outcomes (e.g. distress).
Contributions, limitations, and challenges
Our study has two significant contributions – multi-organizational sampling and the investigation of potential indirect associations between perceived work–family culture and employee well-being. Previous studies have usually concentrated either on non-representative organizational sampling or on direct associations between the work-family culture and outcome variables. Moreover, the well-being indicators we examined were individual-based strain symptoms instead of work attitudes (job satisfaction, commitment, turnover intentions), which have previously received most scrutiny (for a review, see Kinnunen et al., 2005). Furthermore, we were also able to show that a specific form of perceived organizational support (the perceived family-friendliness of an organization) related either directly or indirectly to the well-being of employees, and that this finding was valid in five diverse organizations. Thus, our study provides at least some evidence of criterion validity for the concept of work–family culture. Accordingly, we can conclude that work–family culture is an important theoretical concept, which deserves more attention in the future.
However, certain limitations should also be acknowledged. First, the cross-sectional design of the study did not permit us to test the direction of causality between the phenomena in question. For example, although previous cross-sectional studies have conceived work–family culture as an antecedent for well-being outcomes, this relationship may also work in the opposite direction. Therefore, future longitudinal studies should certainly address this issue. Moreover, a longitudinal design would also allow us to better examine how work–family culture develops in different organizations. The second limitation of the present study concerns sampling. Although our sample was relatively large and diverse, gender segregation was apparent. Clearly, there is a need to study organizational samples in which gender segregation is not so marked. Thirdly, previous quantitative work–family culture studies (including the present study), have excluded family-related outcome variables from their design (e.g. family satisfaction, marital satisfaction). By including such family-related variables in the research design, it would be possible to achieve the integration of organizational, individual, and family level research in a single study.
However, we had two reasons for conducting the analyses at the individual rather than organizational level (i.e. why the level of agreement was not considered). First, when examining the effects of culture or climate dimensions on well-being outcomes at an individual level, it is more important to assess the quality of the culture in question than its strength or consensus, which is a group-level phenomenon (e.g. Lindell & Brandt, 2000). Because this study focused on exploring whether the quality of work–family culture either directly on indirectly relates to employee self-reported distress, we did not examine the association between the strength or consensus of work–family culture with that outcome. Second, level-of-agreement indices and data aggregation continue to be debated issues in climate and culture research, which makes it hard for the researcher to know what level of aggregation (an organization, department or team) is the most appropriate (see Chan, 1998). In our study, where the organizations were quite large, we did not have enough information on departments, which, admittedly, would have formed the most relevant starting-point for data aggregation (i.e. computation of level-of-agreement indices). Department level information would have provided specific knowledge on subcultures prevailing in the organizations and various departments within them. In large organizations, data from which subcultures (and departments) could be identified would provide a more adequate starting-point for data aggregation because there would probably be a higher level of agreement on cultural manifestations within subcultures.
Finally, as Martin (2002) points out, it should be recognized that employees in the same organization (or work unit) may agree (an integration perspective) on some cultural elements or manifestations while disagreeing on others (a differentiation perspective). Applied to work–family culture, this would imply that employees working in the same organization or work unit are likely to share only some perceptions of family-friendliness. Perhaps such a variation cannot be captured by a sophisticated statistical analysis alone (e.g. via level-of-agreement indices), but requires a thorough qualitative research design, which is also an appropriate methodology for work–family culture research (e.g. Lewis & Smithson, 2001).
A major challenge for future research is to examine the associations between work-family culture, work–family conflict, and various outcome variables in different organizations. Such comparative research would yield more accurate knowledge about the predictive validity of work–family culture and conflict/balance measures. It is important to remember that the theoretical debate surrounding the concepts used in research on work–family culture is very recent, which accounts for the lack of a consensus on how to define and measure it (see Behson, 2002; Kinnunen et al., 2005). Clearly, more theoretical work is called for. In addition, the mechanisms by which work–family culture and well-being outcomes are related should be further examined in the future. For example, it is possible that some other factors mediate the effects between work–family culture and self-reported health and strain better than work–family conflict. Therefore, new mediator and moderator factors should also be considered. Given that work–family culture is an organization-based construct, such mediators might be context-specific experiences such as organizational commitment or withdrawal behaviour.
Our results have also clear implications for organizations, even in Finland, where many statutory work–family arrangements exist. In this study, a family–supportive organizational culture included three important cultural dimensions (management, career development and working hours culture), each of which should be improved if the aim is to achieve a family-friendly organizational culture. In order to increase executive responsiveness regarding work–family issues, training and consultation should be provided for managers and supervisors. As a minimum, training should include the dissemination of information about family-friendly measures and behaviour, and the beneficial effects on well-being of a family-friendly atmosphere and arrangements. Second, in order to promote a family-friendly career development culture, HR strategies and policies should be designed on a more long-term basis. Due to rapid changes in the global economy, HR management often functions on a short-term basis (see e.g. Boyd, 2001; Michie & Sheenan-Quinn, 2001), which is likely to have negative implications for an employee's well-being and family life. However, at the individual level, career development plans should be made on a more long-term basis in such a way that breaks/leave from work for family reasons would be allowed.
Third, to change the long-working-hours culture (i.e. the norms, attitudes and behaviour associated with working–time expectations), one important issue should especially be considered. Expectations regarding long working days are deeply rooted in an organization's everyday practices and long working days have become something taken for granted, among the personnel (Schein, 1990, 1999). Thus, raising awareness of the need for cultural change in organizations is clearly a priority. Employees should ask themselves whether long working days are strictly necessary and whether they should be interpreted less as a sign of commitment and more as a sign of poor efficacy (Schein, 1999).
Moreover, because our study confirmed that work–family conflict may mediate the effects between work–family culture and well-being, work–family conflict should be addressed in organizations. For example, HR managers should regularly obtain information (e.g. via surveys, interviews, and through personal discussion) on how their staff succeed in combining work and family demands. If managers and supervisors have no idea what is going on in employees' personal life, then they cannot understand employees' needs or help them to balance their work and personal/family life. Furthermore, it is important that where employees appear to be experiencing work–family conflict, work–family culture should be explored as well. It is possible that a major reason for work–family conflict is the deep-seated nature of unsupportive practices, beliefs, and norms in the organizational identity.
This study was supported by the Finnish Work Environment Fund (grant numbers 100106, 102289) and the Emil Aaltonen Foundation.
Appendix A: Zero-order correlations between the study variables for the public sector organizations
Appendix B: Zero-order correlations between the study variables for the private sector organizations